Depth Estimation Based On Light Field Imaging

The rise of artificial intelligence has led to the booms of various imaging technologies.Light-fields imaging devices have evolved from large-scale array systems to hand-held microlens devices.The updates of hardware devices and the optimization of software technologies are simultaneously driving the field of computer vision.The lightfields camera has attracted wide attention from academia and camera equipment manufacturers because of its unique visual functions,such as focusing after shooting,adjustable depth of focus,and variable views.At the same time,virtual reality and augmented reality have gradually become key problems of the computational photography,light-fields equipment can be used as a reliable data source for many applications,such as 3D reconstruction and industrial measurement.Both the dualcamera technology of HUAWEI's mobile phone and the light-field patent of Apple have demonstrated the importance of the light field in the next generation of imaging applications.The two core issues in the research of light fields include image refocusing and depth estimation of light-fields data.The two studies are essentially identical in application and complementary in function.The research work of the thesis was totally conducted from theory to application.Firstly,the refocusing was implemented as the fundamental research framework of light fields so that we analyzed the characteristics of the light field data.Secondly,A depth estimation algorithm of light-fields data via designing probability measurement models was proposed by us.The specific content are as follows:1.Light-fields refocusing algorithm based on Fourier slice theorem:The thesis introduces the fundamental framework of light field refocusing and analyzes the characteristics of 4D light field data.In combination with the Fourier slice theorem,The refocusing function is implemented by us in the frequency domain where 4D light-fields data is sliced into 2D image data.The process of refocusing provides the research basis and functional verification for the depth estimation of the light-fields data.2.Light-fields depth estimation algorithm based on optimal transmission theory:Following the idea and framework of the light field refocusing,the thesis establishes a relative depth estimation of the light field image by designing probability measurement models on the depth labels which are predefined through disparity.And different depth estimation methods for light fields are analyzed and summarized,then the optimal transmission theory is combined to establish different measurement modes for processing light-fields data in the initial depth calculation period and depth fusion period respectively,thus,difference metrics and similarity metrics.After a series of experiments on different dataset,the depth estimation algorithm proposed in this paper is proved to be robust in the occlusion scenes and noise areas.